Model registry

The model registry is the agent-config building block for model selection: which models agent work runs on per tenant and workload. The registry surface is /ai/models, with Models, Providers, and Gateways tabs and a sync-from-gateway action. Each row carries name, key, provider, capabilities, context window, per-unit price, and status; inactive rows stay visible so deprecations are deliberate.

PRESHos is an operating layer, not a model. It sits between agents and the stack, governing what agents can see, touch, and change; model choice does not change the three guarantees (isolation, approvals, audit).
Provisioning carries an ai_gateway_key step (token-gated and overridable, skippable without blocking core-ready). Provider wiring behind that key is platform-managed; its specifics are not documented in this wiki.
Until this page is sourced: do not promise stakeholders specific models, routing behavior, or per-model capability differences. Check the product UI or ask the PRESHai team.
Whatever the model, the operating discipline holds: one goal per session, scoped environment, gated risky actions, recorded outcome.

Models registry with provider, capabilities, context window, price, and status per model

Related: AI providers for the wiring behind the gateway key. Grounding on the provisioning step: Provisioning and internals.